Evolution of title of long ACL papers. Each column shows terms distinctive a year. Size indicates frequency; order characteristicness.

2017	2018	2019	2020	2021	2022
present	neural	art	bert	language models	existing
word	network	neural	english	graph	tasks
features	words	paper	modeling	datasets	pre - trained
neural	information	dataset	graph	retrieval	languages
algorithm	state	show that	method	training	language
learning	sentiment	used	transformer	based	large
use	introduce	bert	human	generation	plms
problem	parsing	methods	performance	specifically	code
sequence	learning	approach	natural language	translation	evaluation
art	architecture	context	training	text	speech